A neurocomputational theory of how explicit learning bootstraps early procedural learning

Area of Science:

• Cognitive Neuroscience
• Computational Psychology
• Neuroscience of Learning and Memory

Background:

• Human learning and memory rely on multiple interacting systems, including explicit (prefrontal cortex) and procedural (basal ganglia) systems for categorization.
• While distinct systems are suited for different categorization tasks, their precise interaction for optimal learning remains unclear.

Purpose of the Study:

• To investigate the interaction between explicit and procedural memory systems during category learning.
• To identify plausible interaction architectures through computational modeling.

Main Methods:

• Utilized COVIS, a computational model of human category learning incorporating both explicit and procedural systems.
• Explored the model's parameter space across various conditions and architectures for procedurally learned categorization tasks.
• Simulated learning progressions to identify interaction mechanisms.

Main Results:

• Simulation results support a one-way interaction where the explicit system bootstraps learning in the procedural system.
• The procedural system initially adopts a suboptimal strategy from the explicit system, subsequently refining it.
• Potential mechanisms include cortical-striatal projections or explicit system control over basal ganglia-mediated motor responses.

Conclusions:

• The explicit and procedural systems interact dynamically, with the explicit system playing a crucial role in initiating and guiding procedural learning.
• This bootstrapping mechanism may explain how complex category learning is achieved efficiently.
• Findings offer insights into the neural underpinnings of category learning and memory system interactions.